Learning the Physics and Chemistry of Surfaces via Machine Vision and Deep Data Analysis

, , ,

Keywords: , , , , , ,

The last decade has witnessed an enormous increase in the size and quality of datasets produced by microscopic and spectroscopic experimental techniques. Particularly, the recent advances in scanning transmission electron and scanning probe microscopies have opened an unprecedented path towards probing the materials structural parameters and various functional properties in real space with a sub-nanometer precision. Such experimental capabilities require adequate analytical methods for extracting a relevant physical and chemical information from the large datasets in which the information of interest is spatially distributed and has a complex multi-dimensional nature [1]. Here we demonstrate case studies involving our use of machine learning and deep data analysis for reading and recognizing complex molecular, atomic and electronic patterns on surfaces and elucidating physical and chemical mechanisms behind the derived structure-property correlations. First we will present an approach based on synergy of Markov random field model and convolution neural networks for classification of structural and rotational classes of all the individual building blocks in scanning tunneling microscopy data on self-assembly of buckybowls on gold (~1000 molecules per image). We show how the full decoding of molecular states allows us to analyze spatial correlations between multiple order parameters at the nanoscale and elucidate reaction pathway involving molecular conformation changes [2]. In our second case study, we will describe a pathway for direct data mining of structure-function relationships in graphene and high-temperature superconductors. Specifically, we have designed an approach based on a combination of sliding window fast Fourier transform, Pearson correlation matrix and linear and kernel canonical correlation methods to study the relationships between lattice distortions and electronic order parameters such as electron scattering and superconducting gap [3-4]. Our methods represent a significant paradigm shift in our way of analyzing atomic and/or molecular resolved microscopic images and can be applied to variety of other microscopic measurements of structural, electronic, and magnetic orders in different condensed matter systems. References: 1. S. V. Kalinin, B. G. Sumpter and R. K. Archibald. Big-deep-smart data in imaging for guiding materials design. Nature Materials 2015, 14, 973. 2. M. Ziatdinov, A. Maksov, S. V. Kalinin. Learning surface molecular structures via machine vision. Submitted. 3. M. Ziatdinov, A. Maksov, L. Li, A. S. Sefat, P. Maksymovych and S. V. Kalinin. Deep data mining in a real space: separation of intertwined electronic responses in a lightly doped BaFe2As2. Nanotechnology 2016, 27, 475706. 4. M. Ziatdinov, S. Fujii, M. Kiguchi, T. Enoki, S. Jesse and S. V. Kalinin. Data mining graphene: correlative analysis of structure and electronic degrees of freedom in graphenic monolayers with defects. Nanotechnology 2016, 27, 495703.

PDF of paper:

Journal: TechConnect Briefs
Volume: 1, Advanced Materials: TechConnect Briefs 2017
Published: May 14, 2017
Pages: 5 - 8
Industry sector: Advanced Materials & Manufacturing
Topic: Materials Characterization & Imaging
ISBN: 978-0-9975117-8-9